By Michael G. Langan, Murphy J. Foss and Greg R. Wells[1]
On April 10, 2024, the NDNY-FCBA published an article that cited five of the approximately 13 federal court cases that existed at the time in which parties were admonished or sanctioned for filing documents containing inaccuracies created by their use of generative artificial intelligence (“generative AI” or “gen AI”). Since then, the number of such cases has ballooned to at least 147 by the time of publication of this article.[2]
Only 85 of those 147 cases involve the misuse of generative AI by pro se litigants; the remaining 62 cases involve the misuse of generative AI by attorneys. The generative-AI programs used have included ChatGPT, Google Bard, Grok, Claude Sonnet 4, and (to a lesser extent) Lexis+ AI and Westlaw CoCounsel.[3] The court filings in which the misuses have occurred have included complaints, memoranda of law, expert reports, objections to Report-Recommendations, and appellate briefs.
The misuses have included citations to fictitious cases, citations to inapposite cases, citations to fake academic articles, incorrect citations of the authors of academic articles, false quotations from cases, false quotations from a Restatement of the Law, use of AI to respond during a discovery conference, the (requested) use of AI during a court hearing, the use of AI to provide the prevailing hourly attorney rate in Individuals with Disabilities Education Act (IDEA) cases, the use of AI to define a term used in a patent, and the use of AI to provide an overview of a medical disease.
The sanctions have included warnings, the striking of the offending portions of filings, the striking of offending filings in their entirety, the denial of the ability to substitute corrected case citations, the preclusion of evidence, the sending of letters of apology to judges falsely identified as authors, mandatory attendance at a CLE, referrals to district professional misconduct committees, referrals to state bar associations, the award of attorney’s fees or monetary sanctions to the court (ranging from $500 to more than $30,000),[4] the removal from attorneys from the case, the denial of motions, and the dismissal of complaints or appeals.
These 147 cases come from 60 of the 94 federal districts, including two cases from the NDNY.[5] Perhaps most disturbingly, these 147 cases have been joined by two cases in which federal court judicial decisions themselves have been withdrawn and replaced due to the apparent misuse of generative AI by a judicial law clerk.[6]
[1] Murphy J. Foss is a third-year student at George Washington University Law School, and Greg R. Wells is a second-year student at Southern Methodist University Dedman School of Law. Both were judicial interns in the Chambers of United States District Judge Glenn T. Suddaby in the summer of 2025.
[2] A list of citations is available from Michael G. Langan upon request.
[3] See Gonzalez v. Tex. Taxpayers & Rsch. Ass’n, 24-CV-0880, 2025 U.S. Dist. LEXIS 16801, at *1, 4-5 (W.D. Tex. Jan. 29, 2025) (stating that plaintiff’s counsel admitted to using “Lexis Nexis’s AI citation generator” to prepare response brief that contained case hallucinations); Gauthier v. Goodyear Tire & Rubber Co., 23-CV-281, 2024 WL 4882651, at *1 (E.D. Tex. Nov. 25, 2024) (stating that plaintiff’s counsel admitted to using unspecified “generative artificial intelligence tool” to produce response brief that contained case hallucinations, and then used “LexisAI” to try to verify/flag any errors but that the AI failed to pick up the case hallucinations); Williams v. Capital One Bank, N.A., 24-CV-2032, 2025 WL 843285, at *7 (D.D.C. March 18, 2025) (suspecting use of Westlaw CoCounsel based on plaintiff’s citation to “case text” from AI program “CoCounsel” in opposition brief that contained case hallucinations).
[4] See Lacey v. State Farm Gen. Ins. Co., 24-CV-5205, 2025 WL 1363069, at *1, 5 (C.D. Cal. May 5, 2025) (striking supplemental briefs containing nine incorrect legal citations due to use of various generative AI tools, and awarding compensation of $31,100 to defense counsel).
[5] See Brown v. Fat Dough Incorp., 22-CV-0761, 2025 WL 2663170, at *4-5 (N.D.N.Y. Sept. 17, 2025) (Sannes, J.) (cautioning pro se plaintiff that he faces a substantial risk of sanctions if he continues to use artificial intelligence without carefully verifying all citations); Kaur v. Desso, 25-CV-0726, 2025 WL 1895859, at *4 (N.D.N.Y. July 9, 2025) (Nardacci, J.) (sanctioning plaintiff’s counsel $1,000 to be paid to the Clerk of Court, and requiring proof of completion of a CLE-credited program related to the use of artificial intelligence in legal writing).
[6] See Mississippi Assoc. of Educators, et al. v. Bd. of Tr. of State Institutions of Higher Learning, 25-CV-00417, Order Granting Temporary Restraining Order (S.D. Miss. filed July 20, 2025) (copy attached at Dkt. No. 58, Attach. 1, at 3-19 [Exhibit A to motion of July 28, 2025, by Defendants to correct errors in the form of references to non-existent plaintiffs and defendants, citations to non-existent factual allegations, quotations of non-existent statutory language, and reliance upon multiple non-existent declarations]), withdrawn and replaced by 2025 WL 2142676 (S.D. Miss. July 20, 2025), motion for clarification denied, 2025 WL 2206327 (S.D. Miss. Aug. 1, 2025); Levon v. Cormedix, 21-CV-14020, Opinion and Order (D. N.J. filed June 30, 2025) (copy attached at Dkt. No. 123, Attach. 1, at 8-40 [Exhibit A to letter of July 22, 2025, from defense counsel alerting the Court of errors in the form of three misstated case outcomes and numerous fictitious quotations]), withdrawn and replaced by 2025 WL 2400346 (D. N.J. Aug. 19, 2025).